The present invention relates to a warning device, a warning method, and a warning program.
An imaging device such as a digital camera generally takes an image of an area to be imaged by using a filter that transmits light with a specific wavelength. The imaging device can improve the image quality of an image taken by use of a filter that is suitable for the amount of light or the like of an area to be imaged. Thus, in the daytime with a large amount of visible light, a camera is used that employs a filter for daytime which transmits more light in the visible spectrum, and in the nighttime with a small amount of visible light, a camera is used that employs a filter for nighttime which transmits more light in the infrared spectrum.
Japanese Unexamined Patent Application Publication No. H07-160957 discloses a technique related to a monitoring device that includes an infrared sensor and a near-infrared camera. The monitoring device disclosed in Japanese Unexamined Patent Application Publication No. H07-160957 takes an image of an area monitored by the infrared sensor by using the near-infrared camera, and detects a change in the monitored area based on the image taken by the near-infrared camera. The monitoring device then determines an abnormal state based on the detected change and the output of the infrared sensor.
Recently, a warning device has been developed that takes an image of the area around a vehicle by using an in-vehicle camera, performs image recognition of the taken image and detects a person, a vehicle or the like as an object in the image, and issues a warning to a driver. Japanese Unexamined Patent Application Publication No. 2011-146049 discloses a technique related to a human detection device for detecting a person from a taken image. The human detection device disclosed in Japanese Unexamined Patent Application Publication No. 2011-146049 detects the appearance of an object based on an image acquired from an image sensor, determines whether a parameter related to the detected appearance of the object and a comparative parameter related to a human match or not, calculates a human probability indicating the probability that the object is a human according to the degree of matching, and determines that the object is a human when the human probability is equal to or more than a preset human threshold.